Searcharxiv⌕ Search

arXiv subjects

Xiaopeng Zhang

Publications and source records attributed to Xiaopeng Zhang.

At least 19 recordsLinked to original sources
arXiv2026

W2A: Distilling World Priors into Action Models for Real-Time Robotic Control

World Action Models (WAMs) have recently shown great promise in robotic control, benefiting from the future imagination capability of large-scale pretrained video models. However, most existing WAMs require explicitly imagining future states at test time, incurring substantial computational overhead and latency that make them unsuitable for real-time robotic control. In this paper, we propose W2A, a two-stage distillation framework for real-time robotic manipulation that distills intermediate denoising representations from a video diffusion model into an action model during training, while entirely discarding the video model at test time. We hypothesize that although video latent representations are high-dimensional, a compressed low-dimensional subspace is sufficient to provide effective supervision for action generation. This makes it possible to transfer such information effectively into an action model. Extensive experiments in both simulation and real-world settings demonstrate competitive success rates while significantly improving inference efficiency. On LIBERO, W2A achieves a 95.9% average success rate. More importantly, W2A achieves about 21x faster inference than the representative WAM baseline.

cs.RO↗
arXiv2026

Recursive Self-Improvement in Physical AI: Mechanisms, Evidence, and Evaluation

Embodied systems can update policies, world models, skills, and executable programs from deployment experience, allowing one interaction to benefit later tasks. When a system also revises and reuses its data-acquisition, feedback, or learning procedures, experience changes the process that produces capability, introducing recursive self-improvement. This paper surveys research in physical AI through the connection between persistent capability and its improvement process. We first explain the functional distinction between learner updates and improvers, then compare policy learning, imagined experience, memory and skill accumulation, and executable harnesses. These approaches connect to revisions of curricula, rewards, and research programs in different ways. Contact validity, recovery, retention, and resource costs determine whether the resulting gains accumulate during continued deployment. An L0-L3 rubric distinguishes task-local adaptation, persistent capability gain, inherited process revision, and verified improver gain. Representative benchmarks and within-study contrasts relate these distinctions to existing results. We further propose common-data and interactive comparisons that evaluate reference and revised improvers from shared learner starts, separating the use of experience from its acquisition. This perspective connects the diverse methods, physical operating conditions, and evaluation evidence, underpinning embodied RSI. A comprehensive list of studies in this survey is available at https://github.com/nkd-lkz/physical-ai-rsi-survey.

cs.RO↗
arXiv2026

Depth through Recurrence: Towards Ultra-Efficient On-Device ASR

Modern automatic speech recognition (ASR) systems have achieved remarkable accuracy by scaling model depth and capacity, but at the cost of increased memory, energy, and computational demands. On edge devices where ASR is often most needed, such as watches and glasses, deploying such large models is infeasible due to extremely constrained resources. This raises a fundamental question: can we achieve high representational power in deep ASR models without scaling up parameterization? In this work, we revisit the role of depth and identify layer wise representational dynamics, in which most layers learn functionally similar transformations. Motivated by this insight, we propose a block-recurrent ASR architecture that replaces parameterized depth with a small set of recurrent blocks, each consisting of weight-shared layers, thereby preserving effective depth while drastically reducing model size. Through representation-guided grouping and knowledge distillation, block-recurrent models retain the accuracy of large reference models while using an order of magnitude fewer parameters. Across Open ASR Leaderboard benchmarks, a model with only two shared blocks recovers 97.4% of the accuracy of large models on average. The compact backbone further enables efficient on-device personalization through lightweight MoE-LoRA adaptation, allowing user-specific models to match or surpass the accuracy of general-purpose ASR systems with significantly lower inference cost. By reusing weights across recurrent steps, the proposed architecture reduces memory-access overhead and inference energy consumption. Together, these results show that high quality ASR can be achieved without large parameterization, and that structured recurrence provides a principled path towards ultra-compressed, energy-efficient, and personalized speech recognition on the edge.

cs.LG↗
arXiv2026

StackGen3D: Stable 3D Block Stack Generation from Multiview Silhouettes

Generating physically stable 3D structures from visual specifications requires resolving geometric ambiguity while accounting for physical interactions among objects. We present StackGen3D, a framework for generating stable block assemblies conditioned on a given block inventory and three orthographic silhouettes. Our approach consists of two stages. First, we develop a physics-informed graph neural network (PI-GNN) that models support relations between blocks and serves as a fast stability predictor for curating a diverse dataset of stable structures. Second, we train a conditional flow-matching model on this dataset to capture the multimodal distribution of block configurations consistent with the target silhouettes. At inference time, lightweight geometric selection and refinement reduce inter-block penetration without physics simulation. Preliminary experiments show that the PI-GNN generalizes from simple stacks to complex structures substantially better than a parameter-matched multilayer perceptron. For structure generation, Best-of-10 selection improves physical stability from 45.3% to 72.4% while achieving a silhouette IoU of 0.970. These results highlight the potential of combining relational physical reasoning, scalable data curation, and conditional generative modeling for creative 3D assembly. Project website: [https://ripl.github.io/StackGen3D-web/](https://ripl.github.io/StackGen3D-web/)

cs.CV↗
arXiv2026

ORD-WM: Learning Object-Centric Rigid Dynamics for 3D World Modeling in Robotic Manipulation

World models provide robots with a predictive mechanism for reasoning about how actions change object states in manipulation tasks. However, existing image-based and dense 3D world models do not explicitly impose object-level rigid motion, leaving rigid-object geometry preservation to be learned implicitly and making multi-step rollouts susceptible to shape drift or object-level motion inconsistency. To address this, we propose ORD-WM, a 3D world model that learns object-centric rigid dynamics for robotic manipulation. ORD-WM learns a vector-neuron variational autoencoder supervised by signed distance fields, which encodes each object point cloud as a geometry-aware latent representation composed of 3D vector features, and uses an action-conditioned dynamics model to predict object-wise SE(3) motion increments. Rather than generating unconstrained next latent states, ORD-WM analytically applies predicted object-wise rigid transformations to the current latent states, separating persistent object geometry from pose changes. Experiments show that ORD-WM enables stable, geometry-preserving multi-step prediction and supports efficient real-time replanning for closed-loop model predictive control in simulated and real-world manipulation scenarios.

cs.RO↗
arXiv2026

IronViT: Toward Efficient Generalist Visual Representation Learning

A generalist vision encoder must capture semantic, spatial, language-aligned, and action-relevant cues within a unified representation, yet softmax attention underlying today's most capable visual backbones becomes prohibitively expensive at high resolution. A natural attempt to address both challenges is to distill multiple specialist teachers directly into an efficient architecture. We find that directly coupling these objectives degrades representation quality, as the student must simultaneously reconcile heterogeneous capabilities and adapt them to a different token-mixing architecture. We introduce IronViT, built on a simple principle: consolidate capabilities before constraining computation. IronViT first distills complementary specialists into a softmax attention capability bridge, then progressively transfers the consolidated representation to a hybrid softmax-linear attention encoder. A purpose-built data pipeline further curates the distillation corpus for higher information density and broader domain coverage. Across recognition, retrieval, dense prediction, multimodal understanding, and robotic learning, IronViT is competitive with leading specialist and generalist vision encoders. The softmax bridge achieves the strongest aggregate performance in multimodal understanding and robotic learning among the evaluated backbones, while the hybrid encoder retains broad transfer performance with an efficiency advantage that grows with input resolution. Together, these results show that consolidating capabilities before architectural conversion can yield a generalist visual encoder without inheriting the prohibitive high-resolution cost of conventional softmax attention.

cs.CV↗
arXiv2026

DualEraser: Joint Video Object and Effect Removal via Balanced Text-Mask Guidance and Decoupled Locator-Preserver

Video object removal frequently struggles to eliminate target objects and their associated complex physical effects (e.g., smoke and light) in real-world scenes. We attribute this challenge to a fundamental semantic--pixel conflict, which manifests at two aspects: condition-level modality dissonance and optimization-level objective entanglement. In terms of conditioning, modality dissonance emerges from single-modality information incompleteness and cross-modal dominance imbalance. During optimization, two conflicting objectives---high-level semantic erasure and pixel-level background preservation---are inextricably entangled within a single model. To address these conflicts, we propose DualEraser, a novel framework for joint video object and effect removal. First, a Bipartite Text prompt and a Multi-Conditional Capability Elicitation (MCCE) mechanism explicitly inject effect semantics and further leverage multimodal priors to address the limitations of individual modalities. Second, a Learnable Deep CFG Fusion (LD-CFG) module adaptively balances the relative dominance between the text and mask conditions. Finally, we introduce a decoupled expert architecture comprising a Locator for semantic erasure and a Preserver for background pixel alignment to break the objective entanglement. Extensive experiments demonstrate that DualEraser achieves state-of-the-art quantitative performance on standard benchmarks (e.g., gains of 2.16 dB and 1.44 dB on ROSE and VOR-Eval, respectively), while enabling robust removal of complex effects in open-world videos. https://cyqii.github.io/DualEraser.github.io/

cs.CV↗
arXiv2026

DehazeGS: Seeing Through Fog with 3D Gaussian Splatting

Current novel view synthesis methods are typically designed for high-quality and clean input images. However, in foggy scenes, scattering and attenuation can significantly degrade the quality of rendering. Although NeRF-based dehazing approaches have been developed, their reliance on deep fully connected neural networks and per-ray sampling strategies leads to high computational costs. Furthermore, NeRF's implicit representation limits its ability to recover fine-grained details from hazy scenes. To overcome these limitations, we propose learning an explicit Gaussian representation to explain the formation mechanism of foggy images through a physically forward rendering process. Our method, DehazeGS, reconstructs and renders fog-free scenes using only multi-view foggy images as input. Specifically, based on the atmospheric scattering model, we simulate the formation of fog by establishing the transmission function directly onto Gaussian primitives via depth-to-transmission mapping. During training, we jointly learn the atmospheric light and scattering coefficients while optimizing the Gaussian representation of foggy scenes. At inference time, we remove the effects of scattering and attenuation in Gaussian distributions and directly render the scene to obtain dehazed views. Experiments on both real-world and synthetic foggy datasets demonstrate that DehazeGS achieves state-of-the-art performance. visualizations are available at https://jz-y-cn.github.io/DehazeGS/

cs.CV↗
arXiv2026

Learning Robust Execution in Robotic Manipulation with Agentic Reinforcement Learning

Robotic manipulation poses fundamental challenges due to uncertainty, long-horizon execution, and compounding errors, which can easily destabilize execution and lead to task failure. Although recent vision-language-action (VLA) models exhibit strong generalization, they typically lack explicit mechanisms to assess execution stability and to recover when execution deviates from its nominal behavior. In this paper, we propose: (1) two complementary metrics to assess execution quality at runtime, and (2) an agentic reinforcement learning framework that learns to restore effective execution through high-level decision-making rather than directly learning low-level actions. In this framework, an agentic policy reasons over recent execution history and selects among a small set of execution modes to regulate the execution process. Under execution degradation, it triggers appropriate recovery mechanisms to restore the robot to previously visited nominal states, enabling the task to continue. We evaluate the proposed method on the LIBERO benchmark, achieving up to a 13.7% improvement in success rate under standard settings and up to a 39.2% improvement under disturbance settings, demonstrating substantially enhanced execution robustness.

cs.RO↗
arXiv2026

Temporal Action Selection for Action Chunking

Action chunking is a widely adopted approach in Learning from Demonstration (LfD). By modeling multi-step action chunks rather than single-step actions, action chunking significantly enhances modeling capabilities for human expert policies. However, because action chunking makes a single decision only after a complete action block has been executed, the resulting reduction in decision frequency restricts the utilization of real-time observations, impairing reactivity in dynamic or noisy environments. Existing efforts to address this issue have primarily resorted to trading off reactivity against decision consistency, without achieving both. To address this limitation, we propose a novel algorithm, Temporal Action Selection (TAS), which caches predicted action chunks from multiple timesteps and dynamically selects the optimal action through a lightweight selector network. TAS achieves balanced optimization across both reactivity and decision consistency. Experiments across multiple tasks with diverse base policy architectures show that TAS significantly improves success rates. Furthermore, integrating TAS as a base policy with residual reinforcement learning (RL) improves both training efficiency and the performance ceiling. Experiments in both simulation and physical robots confirm the method's efficacy.

cs.RO↗
arXiv2026

Occlusion-Aware Physics-Semantic Keyframe Selection for Robust Video Editing

Video editing has recently achieved remarkable progress with diffusion-based generative models, enabling diverse object-level manipulations from natural language instructions. However, existing methods often struggle under occlusion, viewpoint changes, and fast object motion, where unreliable visual observations lead to inaccurate localization, temporal flickering, and inconsistent edits. In this work, we identify the absence of reliable visual anchors as a fundamental bottleneck in occlusion-robust video editing. To address this issue, we propose an occlusion-aware physics-semantic keyframe selection framework that automatically identifies an optimal anchor frame for downstream editing. Specifically, our method evaluates candidate frames from three complementary perspectives: structural completeness for avoiding truncated observations, cycle-consistent tracking stability for measuring physical reliability, and vision-language-based attribute visibility for ensuring semantic clarity. The selected keyframe is then propagated through bidirectional tracking to generate dense spatiotemporal masks, which are used as auxiliary supervision for a diffusion-based video editing backbone. By transforming occlusion handling from explicit reconstruction into reliable anchor selection, our framework enables precise and temporally consistent editing without requiring manual annotations. Extensive experiments on challenging video editing benchmarks demonstrate the effectiveness and high-quality performance of our method.

cs.CV↗
arXiv2026

Reasoning to Align: Implicit Reasoning in Diffusion Transformers for Video Editing

Instruction-based video editing requires transforming a source video according to a natural-language instruction while preserving irrelevant content and remaining temporally coherent. We argue that existing Diffusion Transformer (DiT) editors struggle with this task for two structural reasons. First, conditioning signals are fed undifferentiated into all transformer blocks, forcing a single token stream to encode both global editing intent and fine-grained visual evidence. Second, the cross-attention patterns that govern the edit are supervised only indirectly through pixel-level reconstruction, leaving the model's internal reasoning process under-constrained. To address both limitations, we propose RVEDiT, an implicit Reasoning Video Editing DiT framework built around two complementary components. The first, Granularity-Routed Token Conditioning, introduces learnable editing tokens distilled from a multimodal LLM and routes them to shallow blocks, while reserving native visual and textual tokens for deeper blocks, thereby inducing a coarse-to-fine editing process inside the backbone. The second, Reference-Anchored Attention Alignment, employs a parameter-sharing reference branch during training and maximizes the mutual information between the attention features of the editing and reference branches, regularizing the model's internal reasoning without incurring any additional inference cost. Experiments on standard instruction-based video editing benchmarks show that RVEDiT consistently outperforms state-of-the-art baselines, with particularly strong gains on localized and compositional edits.

cs.CV↗
arXiv2026

Pelican-Unify 1.0: A Unified Embodied Intelligence Model for Understanding, Reasoning, Imagination and Action

We present Pelican-Unify 1.0, the first embodied foundation model trained according to the principle of unification. Pelican-Unify 1.0 uses a single VLM as a unified understanding module, mapping scenes, instructions, visual contexts, and action histories into a shared semantic space. The same VLM also serves as a unified reasoning module, autoregressively producing task-, action-, and future-oriented chains of thought in a single forward pass and projecting the final hidden state into a dense latent variable. A Unified Future Generator (UFG) then conditions on this latent variable and jointly generates future videos and future actions through two modality-specific output heads within the same denoising process. The language, video, and action losses are all backpropagated into the shared representation, enabling the model to jointly optimize understanding, reasoning, imagination, and action during training, rather than training three isolated expert systems. Experiments demonstrate that unification does not imply compromise. With a single checkpoint, Pelican-Unify 1.0 achieves strong performance across all three capabilities: 64.7 on eight VLM benchmarks, the best among comparable-scale models; 66.03 on WorldArena, ranking first; and 93.5 on RoboTwin, the second-best average among compared action methods. These results show that the unified paradigm succeeds in preserving specialist strength while bringing understanding, reasoning, imagination, and action into one model.

cs.RO↗
arXiv2026

LaplacianFormer:Rethinking Linear Attention with Laplacian Kernel

The quadratic complexity of softmax attention presents a major obstacle for scaling Transformers to high-resolution vision tasks. Existing linear attention variants often replace the softmax with Gaussian kernels to reduce complexity, but such approximations lack theoretical grounding and tend to oversuppress mid-range token interactions. We propose LaplacianFormer, a Transformer variant that employs a Laplacian kernel as a principled alternative to softmax, motivated by empirical observations and theoretical analysis. To address expressiveness degradation under low-rank approximations, we introduce a provably injective feature map that retains fine-grained token information. For efficient computation, we adopt a Nyström approximation of the kernel matrix and solve the resulting system using Newton--Schulz iteration, avoiding costly matrix inversion and SVD. We further develop custom CUDA implementations for both the kernel and solver, enabling high-throughput forward and backward passes suitable for edge deployment. Experiments on ImageNet show that LaplacianFormer achieves strong performance-efficiency trade-offs while improving attention expressiveness.

cs.CV↗
arXiv2026

FineEdit: Fine-Grained Image Edit with Bounding Box Guidance

Diffusion-based image editing models have achieved significant progress in real world applications. However, conventional models typically rely on natural language prompts, which often lack the precision required to localize target objects. Consequently, these models struggle to maintain background consistency due to their global image regeneration paradigm. Recognizing that visual cues provide an intuitive means for users to highlight specific areas of interest, we utilize bounding boxes as guidance to explicitly define the editing target. This approach ensures that the diffusion model can accurately localize the target while preserving background consistency. To achieve this, we propose FineEdit, a multi-level bounding box injection method that enables the model to utilize spatial conditions more effectively. To support this high precision guidance, we present FineEdit-1.2M, a large scale, fine-grained dataset comprising 1.2 million image editing pairs with precise bounding box annotations. Furthermore, we construct a comprehensive benchmark, termed FineEdit-Bench, which includes 1,000 images across 10 subjects to effectively evaluate region based editing capabilities. Evaluations on FineEdit-Bench demonstrate that our model significantly outperforms state-of-the-art open-source models (e.g., Qwen-Image-Edit and LongCat-Image-Edit) in instruction compliance and background preservation. Further assessments on open benchmarks (GEdit and ImgEdit Bench) confirm its superior generalization and robustness.

cs.CV↗
arXiv2026

ParaUni: Enhance Generation in Unified Multimodal Model with Reinforcement-driven Hierarchical Parallel Information Interaction

Unified multimodal models significantly improve visual generation by combining vision-language models (VLMs) with diffusion models. However, existing methods struggle to fully balance sufficient interaction and flexible implementation due to vast representation difference. Considering abundant and hierarchical information in VLM's layers from low-level details to high-level semantics, we propose \textbf{ParaUni}. It extracts features from variants VLM's layers in a \textbf{Para}llel way for comprehensive information interaction and retains a flexible separation architecture to enhance generation in \textbf{Uni}fied multimodal model. Concretely, visual features from all VLM's layers are fed in parallel into a Layer Integration Module (LIM), which efficiently integrates fine-grained details and semantic abstractions and provides the fused representation as a condition to the diffusion model. To further enhance performance, we reveal that these hierarchical layers respond unequally to different rewards in Reinforcement Learning (RL). Crucially, we design a Layer-wise Dynamic Adjustment Mechanism (LDAM) to facilitate multiple reward improvements that aligns the hierarchical properties of these layers using RL. Extensive experiments show ParaUni leverages complementary multi-layer features to substantially improve generation quality and shows strong potential for multiple reward advances during RL stages. Code is available at https://github.com/JosephTiTan/ParaUni.

cs.CV↗
arXiv2026

FineViT: Progressively Unlocking Fine-Grained Perception with Dense Recaptions

While Multimodal Large Language Models (MLLMs) have experienced rapid advancements, their visual encoders frequently remain a performance bottleneck. Conventional CLIP-based encoders struggle with dense spatial tasks due to the loss of visual details caused by low-resolution pretraining and the reliance on noisy, coarse web-crawled image-text pairs. To overcome these limitations, we introduce FineViT, a novel vision encoder specifically designed to unlock fine-grained perception. By replacing coarse web data with dense recaptions, we systematically mitigate information loss through a progressive training paradigm.: first, the encoder is trained from scratch at a high native resolution on billions of global recaptioned image-text pairs, establishing a robust, detail rich semantic foundation. Subsequently, we further enhance its local perception through LLM alignment, utilizing our curated FineCap-450M dataset that comprises over $450$ million high quality local captions. Extensive experiments validate the effectiveness of the progressive strategy. FineViT achieves state-of-the-art zero-shot recognition and retrieval performance, especially in long-context retrieval, and consistently outperforms multimodal visual encoders such as SigLIP2 and Qwen-ViT when integrated into MLLMs. We hope FineViT could serve as a powerful new baseline for fine-grained visual perception.

cs.CV↗
arXiv2026

HMR-1: Hierarchical Massage Robot with Vision-Language-Model for Embodied Healthcare

The rapid advancement of Embodied Intelligence has opened transformative opportunities in healthcare, particularly in physical therapy and rehabilitation. However, critical challenges remain in developing robust embodied healthcare solutions, such as the lack of standardized evaluation benchmarks and the scarcity of open-source multimodal acupoint massage datasets. To address these gaps, we construct MedMassage-12K - a multimodal dataset containing 12,190 images with 174,177 QA pairs, covering diverse lighting conditions and backgrounds. Furthermore, we propose a hierarchical embodied massage framework, which includes a high-level acupoint grounding module and a low-level control module. The high-level acupoint grounding module uses multimodal large language models to understand human language and identify acupoint locations, while the low-level control module provides the planned trajectory. Based on this, we evaluate existing MLLMs and establish a benchmark for embodied massage tasks. Additionally, we fine-tune the Qwen-VL model, demonstrating the framework's effectiveness. Physical experiments further confirm the practical applicability of the framework.Our dataset and code are publicly available at https://github.com/Xiaofeng-Han-Res/HMR-1.

cs.RO↗